Non-linear plant phenotyping pipelines. How can structural models and machine learning can help us analyse large plant image datasets

Many structural root models have been developed, either generic or for specific species, and these have repeatedly been shown to faithfully represent the root system structure, as well as being able to output ground-truthed data for every simulation and image, independent of root system size. Here we will show that structural root models can be used in combination with image analysis pipelines to assess and improve their overall performance. First, we will show that an in-depth analysis of root image analysis pipelines using such models reveals strong limitations in their ability to measure complex root systems. Secondly, we will present an innovative strategy that combines root models and machine-learning algorithms (random-forests), that can increase the measurement accuracy.